mji
Terminal-Bench 4.0
When OpenAI employees have a problem, they email this special address
Snapcompact: SoTA compaction – instant, local, free. Pick 3
The Future of Meta Superintelligence: A 1 Year Progress Update
Sure but that's just because there are now different companies (AI labs) on this list instead
DeepSeek-V4 on Day 0: From Fast Inference to Verified RL with SGLang and Miles
TorchTPU: Running PyTorch Natively on TPUs at Google Scale
RvLLM: High-performance LLM inference in Rust
My chief of staff, Claude Code
Native KV Cache Offloading to Any Filesystem with LLM-D
Mooncake Joins PyTorch Ecosystem
The K-Shaped Future of Software Engineering
If coordination is your limiting factor I'd argue that it shouldn't be, and you're not investing enough in removing it as a factor. Companies can use various tools to do this, for example:
* Defining directly responsible individuals / single-threaded leaders so that every choice doesn't involve massive coordination
* Putting people who work together in the office sitting next to each other most days of the week
* Or, for remote work, having a strong culture of async communication that is visible to the broader group by default, for example with Slack
Now it's 20x at the AI labs instead of 5x at FAANG.
How Prompt Caching Works – Paged Attention and Automatic Prefix Caching
California lawmakers pass SB 79, housing bill that brings dense housing
Reinforcement Learning with Nvidia NeMo-RL
Everyone Mark Zuckerberg Has Hired So Far for Meta's 'Superintelligence' Team
I think this attitude has been taken too far to the point that people (especially senior+ engineers) end up spending massive amounts of time debating and aligning things that can just be done in far less time (especially with AI, but even without it). And these big companies need to change that if they want to get their productivity back. From the article:
> One engineer said that building a feature for the website used to take a few weeks; now it must frequently be done within a few days. He said this is possible only by using A.I. to help automate the coding and by cutting down on meetings with colleagues to solicit feedback and explore alternative ideas."
This sucks for the 50% or so who are like you, but there's another 50% who won't really get much done otherwise, either because they don't know what to do and aren't self-motivated or capable enough to figure it out (common) or because they're actively cheating you and barely working (less common)